Soft Multiuser Decoding for Vector Quantizationover

| An approach to optimal soft decoding for vector quantization over a code-division multiple-access channel is presented. The decoder of the system is soft in the sense that the unquantized outputs of the matched lters are utilized directly for decoding (no decisions are taken), and optimal according to the minimum mean squared error criterion. The derived decoder utilizes a-priori source information and knowledge of the channel characteristics to combat channel noise and multiuser interference in an optimal fashion. Hadamard transform representations for the user vector quantizers are employed in the derivation and for the implementation of the decoder. The advantages of this approach are emphasized. Suboptimal versions of the optimal decoder are also considered. Simulations show the soft decoders to outperform decoding based on maximum likelihood multiuser detection. Furthermore, the suboptimal versions are demonstrated to perform close to the optimal, at a signiicantly lower complexity in the number of users. The introduced decoders are, moreover, shown to exhibit near-far resistance. Simulations also demonstrate that combined source-channel encoding, with joint source-channel and multiuser decoding, can signiicantly outperform a tandem source-channel coding scheme employing multiuser detection plus table look-up source decoding.